Track 2: AI-Powered Document Analysis & Extraction
1. Problem Statement
Create an intelligent document processing system that can extract, analyse, and summarise
content from various document formats (PDF, DOCX, image with text). The system must
leverage AI to understand document structure and extract key information automatically.
2. Key Features
● Multi-format support: PDF, DOCX, and image (via OCR)
● Automatic text extraction with layout preservation
● AI-powered summarisation
● Key entity extraction: names, dates, organisations, monetary amounts etc.
● Sentiment analysis for text content
3. Technical Requirements
● OCR: Tesseract / Google Cloud Vision API
● AI/ML: Flexible model for summarisation
● Backend: Any language Celery for async processing
4. Submission Requirements
● Live deployed URL (must be publicly accessible)
● API Endpoint
● GitHub repository link
5. REST API Design
● The API should accept one document at a time.
● Supported formats: PDF, DOCX, or image (via OCR).
● The document will be sent as a Base64 encoded string in the request body.
● The system should analyze the document content.
● The API must return the extracted key information in JSON format.
6. API Authentication
Requests without a valid API key must be rejected with a 401 Unauthorized status.
● Header Format: x-api-key: YOUR_SECRET_API_KEY
7. API Request (cURL Example)
Endpoint Example: POST [Link]
cURL Request:
curl -X POST [Link] \
-H "Content-Type: application/json" \
-H "x-api-key: sk_track2_987654321" \
-d '{
"fileName": "[Link]",
"fileType": "pdf",
"fileBase64": "JVBERi0xLjQKJcfsj6IKNSAwIG9iago8PC9UeXBlIC9QYWdl..."
}'
8. Request Body Fields
Field Description
fileName Uploaded file Name
fileType pdf/docx/image
fileBase64 Base64-encoded file
9. API Response Body (Success)
Example Response
"status": "success",
"fileName": "[Link]",
"summary": "This document is an invoice issued by ABC Pvt Ltd to Ravi Kumar on 10 March
2026 for an amount of ₹10,000.",
"entities": {
"names": ["Ravi Kumar"],
"dates": ["10 March 2026"],
"organizations": ["ABC Pvt Ltd"],
"amounts": ["₹10,000"]
},
"sentiment": "Neutral"
10. Response Field Explanation
Field Meaning
status success or error
fileName Uploaded file Name
summary AI-generated summary of the
document content.
entities Key information extracted from the
document.
sentiment Positive / Neutral /
Negative
11. GitHub Repository Requirements
your-repo/
├── [Link] # Setup and usage instructions
├── src/ # Source code
│ ├── [Link]
├── [Link] # Python dependencies
├── .[Link] # Environment variables template
12. Minimum README Content
# Data Extraction API
## Description
Brief description of your approach and strategy
## Tech Stack
- Language/Framework
- Key libraries
- LLM/AI models used (if any)
## Setup Instructions
1. Clone the repository
2. Install dependencies
3. Set environment variables
4. Run the application## Approach
Explain your Data Extraction strategy:
- How you extract summary, entities and analyze sentiment
13. What We Look For in Code Review
✅ Acceptable Practices:
● Using LLMs/AI models for data extraction and analysis
● Third-party APIs for enhanced detection
❌ Unacceptable Practices:
● Hardcoded responses specific to test scenarios
● Pre-mapped answers based on known test data
Code Review Impact:
● ✅ Passes Review: Score remains unchanged, submission valid
● ⚠️Minor Issues: Warning issued, score may be adjusted
● ❌ Fails Review: Disqualification from hackathon, score set to 0
14. Scoring Rubric — 100 Points
1. API Functionality & Accuracy — 90 Points
The API will be evaluated using 15 test cases consisting of 5 PDF files, 5 DOCX files,
and 5 images containing text. Each file will be sent to the API for processing, and the
response will be analyzed based on the quality and accuracy of the extracted
information.
Each test case carries 10 points, divided into the following components:
● Summary — 2 points
The API should generate a concise and accurate summary of the content.
● Entities — 4 points
The API should correctly extract named entities such as people, organizations,
locations, or other relevant entities present in the text.
● Sentiment — 4 points
The API should correctly classify the overall sentiment of the content as positive,
negative, or neutral.
Since there are 15 test cases, the maximum raw score is 150 points. This raw score
will then be scaled down to a final score out of 90 points for the API functionality and
accuracy evaluation.
Final Score Formula
Final Score = (Total Score from 15 Tests / 150) × 90
2. GitHub Repository Code Quality —10 Points
● Code structure and readability
● Features & Functionality
● Technical Implementation